New Network Design - Advice needed

cisconoobie · ‎07-24-2007

There will be 3 floors. 2 floors will have 1 4510 and 1 floor will have 2 4510's plus 3750g's.

I need to make a redundant network that will provide the best throughput to the servers, internet and VOIP.

The POE blades will be used only for VOIP and the non poe blades are either for servers or clients.

Can you please tell me which diagram is better?

I feel that diagram A allows a better collapsed core since it will be doing only the layer 3 routing. But what makes me wonder is that I am wasting the 4510's and only using them as access switches.

Diagram B seems like a good choice except I will have clients on the 4510's and makes me feel that the 2 4510's will be over subscribed by the clients.

Any issues anyone can see from these diagrams?

What does everyone think?

Amit Singh · ‎07-24-2007

I would rather go for Design B. This is because you will have 3750 switch dedicated for Servers and you will be using L3 links to connect to the core which will give you better redundancy and load-balancing. Also, you server will get connect to 3750 switches which will provide non-blocking architecture (depending upon the model number used and server connectivity) and no oversubscription of the traffic.

In design A, your server will get connect to 4500 directly which will be oversubscribed connectivity as 4500 line card slots doesnot proivde line rate connectivity to backplane.

If I were you, I would have used design B with L3 links between all the 4500's. I will not even run HSRP, as this will extend my L2 domain to switches acting as core and my core has to process unncessary L2 broadcast traffic, STP traffic etc. I would rather use L3 links and run a routing protocol which would give me quicker convergence in case of a link failure, automatic load-balancing. I would create local vlans on each floor and do the inter-vlan routing locally on each 4500 for the hosts.

These were just my inputs on this.

HTH,Please rate if it does.

-amit singh

cisconoobie · ‎07-24-2007

I thought about that Amit but then I would not be able to Span Vlans.

If I would use Diagram B and setup Layer 3 vlans on each 4510. Would the 4510's in the middle acting as Vlan gateways for the directly connecting clients cause contention or oversubscription. Since these 2 4510's are also acting as the Core routing packets from all switches?

Also for VoIP, I would have to create multiple voice vlans?

Amit Singh · ‎07-24-2007

Yes you would not be able to SPAN the vlans across the entire network. But Why would you want to do that. That will carry the unneccsary L2 traffic to the core.

Clients connecting to each 4510R will not face much of the problem in this case as most of the traffic will be routed across the 4500 and Servers. You also have to check the traffic pattern in the network. If your majority of the traffic is from Clients to the servers on the network, then you might see a lots of performance issue in Design A, as the servers will be using the shared backplane switching connections on the 4510 along with the cleint traffic. If majority of the traffic is to the servers its always advisable to have dedicated server farm switches.

Regarding the voice vlan concern, you can have at max 254 ip phones/hosts in a vlan. If you host count is more than 254 on each floor, then you might have to go for multiple vlans anyways. Moreover, this will give you another Layer2 security on voice vlan as the users on one 4510 chassis will not be able to eavsdrop the voice traffic running on another chassis.

HTH,Please rate if it does.

-amit singh

mchoo2005 · ‎07-24-2007

I think Diagram A is better. It provides you with a clearer architecture of collapsed core/distribution using the 2x 3750G. If possible, try to use L3 links as much as possible.

94halldah · ‎07-25-2007

I agree, you should use L3 links as much as possible. I don't see a need to ether channel L3 links. The routes would be equal cost load balanced, so you are not gaining bandwidth and each path would be in the route table so you are not gaining any performance on route convergence.